feat: slim core install — move unused ML stack to optional extras

[rosspeili]

Problem Statement

requirements.txt pins chromadb, sentence-transformers, and torch, but no module under rooms/ or cli.py imports them. Every pip install -r requirements.txt downloads a large ML stack before a user can run a single CLI session.

That hurts first-time onboarding, slows CI, and conflicts with the project's local-first, lightweight CLI story—especially when vector memory (#13) is not implemented yet.

Proposed Solution

• Split dependencies into a core set (what Rooms uses today: litellm, rich, pydantic, pyyaml, prompt_toolkit, etc.) and an optional extra for future memory/RAG work (chromadb, sentence-transformers, torch).
• Prefer one clear install path, e.g. pip install -r requirements.txt for core only, plus requirements-memory.txt or pip install -e ".[memory]" if using pyproject.toml.
• Update README Quick Start and CI (/.github/workflows/ci.yml) to install core only.
• Document when and why to install the memory extra (link to feat: long-term agent memory across sessions (sidecar storage) #13 when relevant).

Alternatives Considered

• Leave the monolithic requirements.txt — rejected; punishes every user and contributor for unused packages.
• Remove ML packages entirely with no extras — rejected; feat: long-term agent memory across sessions (sidecar storage) #13 and future memory features will need a documented optional path.

Priority

High 🔴

Additional Context

Complements the settings work in #28. Independent of #31, #29, and #8. Good first issue if scoped to requirements + README + CI only.

[AaineeSinha]

Hi! I'd love to take this on as my next issue. Could you please assign it to me?

[AaineeSinha]

I have submitted a pull request to resolve this issue! It splits the dependencies into core and optional extras and updates the README documentation accordingly. Ready for review whenever you have time.